As a paying ChatGPT Plus user, being hit with a hard weekly lockout completely breaks the value proposition of the subscription. When we exhaust our higher-tier allocations or rate limits, Codex should gracefully step down to an included base/fallback tier. We understand that unlimited compute isn’t viable, but locking a paid subscriber out for days is unacceptable—any secondary fallback limit should run on a short rolling window (e.g., 5-hour reset) instead of a weekly hard cap.
Key Pain Points as a Paid Subscriber
-
The Multi-Day Deadlock on a Paid Plan:
One heavy day of sprint debugging or architectural refactoring early in the billing cycle can drain our allowance, completely bricking the tool for the rest of the week. Paying subscribers cannot afford to wait 4 to 6 days for core tooling to become usable again.
-
No Value During Downtime:
We understand server constraints during peak loads, but a weekly lockout treats Plus users like free-tier abuse cases. A short rolling window protects platform infrastructure while ensuring our paid subscription remains functional across the entire workweek.
Proposed Model: Two-Tier Sliding Allowance for Plus
| Tier / Feature | Allocation Type | Limit Structure | Plus User Workflow Impact |
|---|---|---|---|
| High-Tier / Primary Models | Deep reasoning, complex refactors | Plan-based allowance / quota | Preserved for high-value engineering tasks |
| Included Fallback Tier | Routine completions, quick edits | Rolling 5-hour window (NO weekly cap) | Continuous daily utility; work never stalls for days |
Specific Recommendations
-
Keep the Fallback on a 5-Hour Rolling Reset:
Base the lower-tier fallback model strictly on a short rolling cooldown. Even if a subscriber burns through their allocation during an intense morning session, the tool recovers by the afternoon instead of remaining locked until next Monday.
-
Automated Step-Down Instead of Hard Lockout:
When primary credits run dry, notify the user and fail over smoothly: “High-tier quota depleted. Continuing on Fallback Mode (resets in 5h).”
-
Partitioned Workload Meters:
Don’t let background repo sweeps or agentic runs silently exhaust the baseline quota needed for day-to-day inline coding and syntax fixes.